Electron beam deposition energy prediction method and system based on convolution prediction model

Abstract

The invention discloses an electron beam deposition energy prediction method and system based on a convolution prediction model, and the method comprises the steps: S1, collecting two-dimensional plane deposition energy data of beam spots of various sizes through Monte Carlo simulation, carrying out the abnormal data recognition and abnormal data replacement, and obtaining the final two-dimensional plane deposition energy data; s2, obtaining a binarized graph function corresponding to each beam spot, and performing convolution with the binarized graph function corresponding to each beam spot by using a sedimentary energy density function to obtain a corresponding convolution prediction model; s3, fitting the deposition energy density function based on the final two-dimensional plane deposition energy data to obtain an optimal fitting value of each parameter; and S4, substituting the optimal fitting value of each parameter into the deposition energy convolution prediction model corresponding to the beam spot in any shape so as to realize deposition energy prediction of the beam spot in any shape. The method has the advantages of high prediction precision, small calculation amount and the like.

Description

Technical Field

[0001] This invention relates to the field of electron beam technology, specifically to an electron beam deposition energy prediction method and system based on a convolution prediction model. Background Technology

[0002] In the field of chip manufacturing, electron beam lithography offers higher precision compared to photolithography, and is therefore widely used in chip research and development, photomask manufacturing, and other areas with high precision requirements. During exposure, the electron beam interacts with the photoresist, generating deposition energy within the photoresist. The distribution of this deposition energy is a direct factor affecting the exposure accuracy of the pattern and has a significant impact on the exposure results. However, during exposure, electrons scatter around the exposed area, causing the deposition energy to diffuse and creating a proximity effect, which in turn affects the exposure accuracy. Therefore, to correct the exposed pattern, it is necessary to study the deposition energy distribution and achieve deposition energy prediction.

[0003] Given the distribution of the speckle pattern, depositional energy can be predicted by constructing a depositional energy convolution prediction model. First, a depositional energy density function (DEDF) is constructed based on the point spread function (PSF). Then, based on a two-dimensional planar convolution formula, the depositional energy density function is convolved with the binarized speckle pattern function to obtain the convolution prediction model. The input to the convolution prediction model is the location coordinates and speckle dose; the output is the depositional energy per unit volume at that location. The prediction accuracy of the convolution prediction model depends on the parameters of the depositional energy density function; therefore, fitting these parameters is crucial.

[0004] Algorithms that can be used to fit the parameters of the deposition energy density function include numerical analysis methods, single-point search algorithms, and swarm intelligence search algorithms. Convolutional prediction models have complex expressions, making numerical analysis methods difficult to apply; single-point search algorithms are inefficient and prone to getting trapped in local optima; swarm intelligence algorithms offer good parameter fitting results, but computationally intensive tasks are required when fitting the parameters of the deposition energy density function for large-sized speckles.

[0005] One fitting method is to start from the center of the beam spot and statistically analyze the deposition energy density at different distances along a certain direction for parameter fitting. However, this method is easily affected by the shape and size of the beam spot, resulting in low accuracy of the fitted parameters. If the error between the output of the convolution prediction model and the actual deposition energy is used to fit the deposition energy density function parameters, the following problems are faced: (1) The deposition energy data of two-dimensional plane electron beam exposure has random fluctuations and may have abnormal data; (2) When the beam spot size is small, the fitted deposition energy density function parameters may have errors and cannot be applied to the convolution prediction model of other beam spots; while when the size is large, the convolution operation will be very large, resulting in a slow fitting speed, and therefore a long calculation time is required.

[0006] Anomaly processing in sedimentary energy data includes anomaly identification and anomaly replacement. Common algorithms for anomaly identification include the three-fold variance algorithm, deep learning algorithms, and the Local Outlier Factor (LOF) algorithm. The three-fold variance algorithm has relatively low computational cost but low accuracy; deep learning algorithms have high computational cost and poor interpretability; the LOF algorithm has good identification performance and interpretability, but considering the large amount of two-dimensional planar sedimentary energy data, performing anomaly identification for each data sample would be too computationally intensive, thus requiring improvement. Common methods for anomaly replacement include filtering methods, such as mean filtering and Gaussian filtering. However, after anomaly identification, data needs to be removed, which results in missing data, leading to less filtered data and making the anomaly replacement results more susceptible to random fluctuations. Therefore, improvements to the filtering algorithms are needed. Summary of the Invention

[0007] To address the technical problems existing in the prior art, this invention provides an electron beam deposition energy prediction method and system based on a convolutional prediction model, which has high prediction accuracy and low computational cost.

[0008] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows: An electron beam deposition energy prediction method based on a convolutional prediction model includes the following steps: S1. Two-dimensional planar deposition energy data of various beam sizes were collected by Monte Carlo simulation, and abnormal data were identified based on the extreme value-LOF algorithm and abnormal data were replaced based on symmetric filtering to obtain the final two-dimensional planar deposition energy data. S2. Obtain the binarized pattern function corresponding to each spot, and convolve the deposition energy density function with the binarized pattern function corresponding to each spot to obtain the corresponding convolution prediction model. S3. Based on the final two-dimensional planar deposition energy data obtained in step S1, the deposition energy density function is fitted to obtain the optimal fitting values ​​for each parameter. S4. Substitute the optimal fitting values ​​of each parameter into the deposition energy convolution prediction model corresponding to the arbitrary shape of the beam spot to achieve deposition energy prediction for the arbitrary shape of the beam spot.

[0009] Preferably, the specific process of step S1 is as follows: S101. Deposition Energy Data Acquisition: Using Monte Carlo simulation software, the two-dimensional plane is divided into multiple square grids of the same size, with each grid representing a pixel. Simulations are then performed under the same process conditions for exposure of 1x1, 10x10, 5x20, and 100x100 rectangular beams, acquiring deposition energy data within each grid of the two-dimensional plane. The 1x1 beam is referred to as the pixel beam; 10x10 and 5x20 beams are considered small beams; and the 100x100 beam is considered a large beam. S102. Anomaly identification based on the extreme value-LOF algorithm: After completing the two-dimensional planar deposition energy data acquisition, for the deposition energy generated by each type of spot, firstly, the deposition energy data in each grid is traversed to perform 8-neighborhood extreme value detection, including maximum value detection and minimum value detection; then, for the detected extreme value data, other data in its 5x5 neighborhood are used to perform anomaly identification based on the LOF algorithm to determine whether the extreme value data is anomaly data; S103. Abnormal data replacement based on symmetry filtering: For each identified abnormal data, firstly, other deposition energy data within its surrounding 5x5 neighborhood are selected, and the abnormal data is removed. The remaining data is then subjected to mean filtering to obtain the neighborhood filter value. Subsequently, the symmetric coordinates of the abnormal data coordinates about each symmetry axis of the beam spot are calculated, and the neighborhood filter value of each corresponding grid is calculated. Finally, the average value of all neighborhood filter values ​​is calculated as the final filtering result, which is used for abnormal data replacement.

[0010] Preferably, the specific process of step S2 is as follows: S201. Binarization of graphic data: For each spot, set the value of the area covered by the spot in the two-dimensional plane to 1 and the value of the other areas to 0 to obtain the binarized graphic function corresponding to each spot. S202. Construct a deposition energy density function DEDF based on the selected point diffusion function PSF, which is used to calculate the deposition energy generated by electrons in each grid on the surrounding grid. S203. Based on the two-dimensional planar convolution formula, the deposition energy density function is convolved with the binarized graphic function corresponding to each spot to obtain the corresponding convolution prediction model.

[0011] Preferably, the deposition energy density function DEDF The expression is as follows:

[0012] in K This is the proportionality coefficient. D The beam spot dose is the parameter to be fitted. K as well as PSF The parameter (x, y) is the position coordinate.

[0013] Preferably, the specific process of step S3 is as follows: S301. Parameter initialization based on pixel spot deposition energy distribution data and spot dose obtained in step S1: First, calculate the scaling factor K in the deposition energy density function using the pixel spot deposition energy distribution data and spot dose. Then, use the particle swarm optimization algorithm to initialize the parameters in the deposition energy density function. PSF The parameters are fitted to obtain the initial values ​​of the PSF parameters. Where N is the number of PSF parameters; S302. Based on the deposition energy data of the two small-sized spot sizes obtained in S1, the PSF parameter in the deposition energy density function is fitted to obtain the corrected values ​​of the PSF parameter in the deposition energy density function. S303. Using the deposition energy data of large-size beams obtained in S1, the optimal parameters are searched iteratively using a bisection method based on the initial and corrected values ​​of the PSF parameters to obtain the optimal fitting values ​​of each parameter.

[0014] Preferably, the specific process of step S302 is as follows: initialize the PSF parameter initial value obtained from parameter initialization. Adding or subtracting 50% is used as the upper and lower limits for parameter correction. Then, using the deposition energy data of the two small-sized spot sizes obtained in S1, the PSF parameter in the deposition energy density function is fitted based on the particle swarm optimization algorithm within the upper and lower limits of parameter correction, and the corrected values ​​of the PSF parameter in the deposition energy density function are obtained respectively. (i=1,2).

[0015] Preferably, the specific process of step S303 is as follows: using the deposition energy data of the large-size spot obtained in S1, the average gradient of PSF parameter variation is calculated based on the two sets of PSF parameter correction values ​​and initial values. , and then with With the origin, along The direction is searched iteratively using a binary search method, with an initial step size of . The iterative search stops when the change in the objective function is less than the threshold in each search; the final search result is the optimal fit value of the PSF parameter in the deposition energy density function.

[0016] Preferably, the proportionality coefficient K is calculated as follows: .

[0017] Given a 2D plane with a grid of M pixels, where each pixel's coordinates are (x, y, y). i ,y i ), with an area of ​​S, and the deposition energy within each pixel grid is E(x). i ,y i The pixel spot dose is D0.

[0018] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, the computer program performing the steps of the method described above when run by a processor.

[0019] The present invention further discloses an electron beam deposition energy prediction system based on a convolution prediction model, comprising an interconnected memory and a processor, wherein the memory stores a computer program, and the computer program executes the steps of the method described above when run by the processor.

[0020] Compared with the prior art, the advantages of the present invention are as follows: This invention proposes an anomaly identification method based on the extreme value-LOF algorithm to address the characteristics of sedimentation energy anomaly data. Compared with directly using the LOF algorithm, this method reduces the computational load while ensuring the accuracy of anomaly identification. To address the impact of data loss caused by removing anomaly data on anomaly data replacement, this invention proposes an anomaly data replacement method based on symmetric filtering, which increases the amount of filtered data and reduces the interference of random fluctuations on anomaly data replacement.

[0021] This invention directly uses the error between the output of the convolutional prediction model and the actual deposition energy to fit the parameters of the deposition energy density function. Compared with using deposition energy data at different distances from the center of the spot for parameter fitting, this improves the fitting accuracy.

[0022] This invention employs the particle swarm optimization (PSO) algorithm, a swarm intelligence search algorithm, to fit the energy density function parameters of small-sized speckle deposits. Then, using the fitting results, it applies other computationally less computationally intensive methods (such as gradient bisection) to fit the energy density function parameters of large-sized speckles. Specifically, the gradient bisection method is used for fitting the deposition energy density function parameters. Utilizing large-sized speckle deposition energy data, and comprehensively considering gradient information and parameter correction values ​​for the deposition energy density function, the method fits the parameters based on the bisection principle. This significantly reduces computational cost compared to directly using a swarm intelligence search algorithm while ensuring accuracy. Furthermore, in the parameter correction process based on small-sized speckles, this invention uses the ±50% neighborhood of the parameter initialization result obtained in the previous step as the parameter search range, which increases search accuracy and reduces unnecessary computational overhead. Attached Figure Description

[0023] Figure 1This is a flowchart of an embodiment of the electron beam deposition energy prediction method of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0025] like Figure 1 As shown, the electron beam deposition energy prediction method based on a convolutional prediction model provided in this embodiment of the invention includes the following steps: S1. Abnormal data identification and replacement, the specific steps are as follows: S101. Deposition Energy Data Acquisition. Using Monte Carlo simulation software, the two-dimensional plane is divided into multiple square grids of the same size, each grid representing a pixel. Simulations are then performed under the same process conditions for exposure of 1x1, 10x10, 5x20, and 100x100 rectangular beams, acquiring deposition energy data within each grid of the two-dimensional plane. The 1x1 beam, referred to as the pixel beam, is mandatory; 10x10 and 5x20 beams are considered small beams; and the 100x100 beam is considered a large beam. The small and large beams can be replaced with beams of similar size.

[0026] S102. Anomaly identification based on the extreme value-LOF algorithm. After completing the acquisition of two-dimensional planar deposition energy data, for the deposition energy generated by each type of spot, firstly, 8-neighborhood extreme value detection is performed on the deposition energy data in each grid, including maximum and minimum value detection. Then, for the detected extreme value data, anomaly identification based on the LOF algorithm is performed using other data in its 5x5 neighborhood to determine whether the extreme value data is abnormal.

[0027] S103. Anomalous Data Replacement Based on Symmetry Filtering. For each identified anomalous data point, firstly, other depositional energy data within a 5x5 neighborhood are selected, and the anomalous data is removed. The remaining data is then subjected to mean filtering to obtain a neighborhood filter value. Next, the symmetric coordinates of the anomalous data coordinates about each axis of symmetry of the beam spot are calculated, and the neighborhood filter value for each corresponding grid is calculated using the method described above. Finally, the average of all the above neighborhood filter values ​​is calculated as the final filtering result, which is used for anomalous data replacement.

[0028] S2. Construction of the sedimentation energy convolution prediction model, the specific steps are as follows: S201. Binarization of graphic data. For each spot, set the value of the area covered by the spot in the two-dimensional plane to 1, and the value of other areas to 0, to obtain the binarized graphic function corresponding to each spot.

[0029] S202. First, based on the selected point diffusion function PSF, a deposition energy density function DEDF is constructed to calculate the deposition energy generated by electrons in each grid on the surrounding grid.

[0030] The expression for the deposition energy density function is as follows:

[0031] Where K is the proportionality coefficient, D is the beam spot dose, the parameters to be fitted are K and the parameters in PSF, and (x,y) are the position coordinates.

[0032] S203. Then, based on the two-dimensional plane convolution formula, the deposition energy density function is convolved with the binarized graphic function corresponding to each spot to obtain the corresponding convolution prediction model.

[0033] The input to the convolutional prediction model is the position coordinates (x, y) of any grid in a two-dimensional plane and the beam spot dose D. The output is the density of the deposited energy generated by the beam spot of this type in the corresponding grid.

[0034] S3. Parameter fitting based on gradient bisection method. In this step, when searching and fitting the parameters of the deposition energy density function, the target loss function is the average absolute error between the predicted and actual values ​​of the deposition energy density output by the convolutional prediction model.

[0035] S301. Parameter initialization based on pixel spot deposition energy distribution data. First, the proportional coefficient K in the deposition energy density function is calculated using the pixel spot deposition energy distribution data and spot dose obtained in step S1. Then, the PSF parameter in the deposition energy density function is fitted using the particle swarm optimization algorithm to obtain the initial value of the PSF parameter. , where N is the number of PSF parameters.

[0036] The scaling factor K is calculated as follows: Assume the two-dimensional plane has a grid of M pixels, and the coordinates of each pixel are (x, y, y). i ,y i ), with an area of ​​S, and the deposition energy within each pixel grid is E(x). i ,y i If the pixel spot dose is D0, then the formula for calculating K is as follows:

[0037] S302. Parameter correction based on small beam size. First, initialize the PSF parameters obtained from parameter initialization. Adding or subtracting 50% is used as the upper and lower limits for parameter correction. Then, using the deposition energy data of the two small-sized spot sizes obtained in S1, the PSF parameter in the deposition energy density function is fitted based on the particle swarm optimization algorithm within the upper and lower limits of parameter correction, and the corrected values ​​of the PSF parameter in the deposition energy density function are obtained respectively. (i=1,2).

[0038] S302. Precise parameter fitting based on large-size beams. First, using the deposition energy data of the large-size beams obtained in S1, the average gradient of PSF parameter variation is calculated based on two sets of PSF parameter correction values ​​and initial values. , and then with With the origin, along The direction is searched iteratively using a binary search method, with an initial step size of . The iterative search stops when the change in the objective function (the change in the average absolute error between the predicted and actual values ​​of large-size spot deposition energy) is less than the threshold. The final search result is the accurate fit of the PSF parameter in the deposition energy density function.

[0039] S4. After completing the parameter fitting of the deposition energy density function, substitute the parameters into the deposition energy convolution prediction model corresponding to the arbitrary shape of the spot, thereby realizing the deposition energy prediction for the arbitrary shape of the spot.

[0040] This invention proposes an anomaly identification method based on the extreme value-LOF algorithm to address the characteristics of sedimentation energy anomaly data. Compared with directly using the LOF algorithm, this method reduces the computational load while ensuring the accuracy of anomaly identification. To address the impact of data loss caused by removing anomaly data on anomaly data replacement, this invention proposes an anomaly data replacement method based on symmetric filtering, which increases the amount of filtered data and reduces the interference of random fluctuations on anomaly data replacement.

[0041] This invention directly uses the error between the output of the convolutional prediction model and the actual deposition energy to fit the parameters of the deposition energy density function. Compared with using deposition energy data at different distances from the center of the spot for parameter fitting, this improves the fitting accuracy.

[0042] This invention employs the particle swarm optimization (PSO) algorithm, a swarm intelligence search algorithm, to fit the energy density function parameters of small-sized speckle deposits. Then, using the fitting results, it applies other computationally less computationally intensive methods (such as gradient bisection) to fit the energy density function parameters of large-sized speckles. Specifically, the gradient bisection method is used for fitting the deposition energy density function parameters. Utilizing large-sized speckle deposition energy data, and comprehensively considering gradient information and parameter correction values ​​for the deposition energy density function, the method fits the parameters based on the bisection principle. This significantly reduces computational cost compared to directly using a swarm intelligence search algorithm while ensuring accuracy. Furthermore, in the parameter correction process based on small-sized speckles, this invention uses the ±50% neighborhood of the parameter initialization result obtained in the previous step as the parameter search range, which increases search accuracy and reduces unnecessary computational overhead.

[0043] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when run by a processor, executes the steps of the method described above. This invention further provides an electron beam deposition energy prediction system based on a convolutional prediction model, including an interconnected memory and a processor, wherein the memory stores a computer program that, when run by a processor, executes the steps of the method described above. The medium and system of this invention, corresponding to the methods described above, also possess the advantages described above.

[0044] The present invention can implement all or part of the processes in the methods of the above embodiments, or it can be implemented by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium includes: any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. The memory is used to store computer programs and / or modules. The processor implements various functions by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory. The memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0045] Explanation of related terms: PSF (Point Spread Function): Point spread function; DEDF (Deposited Energy Density Function): Deposited energy density function; LOF (Local Outlier Factor): Local outlier factor algorithm.

[0046] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for predicting electron beam deposition energy based on a convolution prediction model, characterized in that, Including the following steps: S1. Two-dimensional planar deposition energy data of various beam sizes were collected by Monte Carlo simulation, and abnormal data were identified based on the extreme value-LOF algorithm and abnormal data were replaced based on symmetric filtering to obtain the final two-dimensional planar deposition energy data. S2. Obtain the binarized pattern function corresponding to each spot, and convolve the deposition energy density function with the binarized pattern function corresponding to each spot to obtain the corresponding convolution prediction model. S3. Based on the final two-dimensional planar deposition energy data obtained in step S1, the deposition energy density function is fitted to obtain the optimal fitting values ​​for each parameter. S4. Substitute the optimal fitting values ​​of each parameter into the deposition energy convolution prediction model corresponding to the arbitrary shape of the beam spot to achieve deposition energy prediction for the arbitrary shape of the beam spot.

2. The electron beam deposition energy prediction method based on a convolution prediction model according to claim 1, characterized in that, The specific process of step S1 is as follows: S101. Deposition Energy Data Acquisition: Using Monte Carlo simulation software, the two-dimensional plane is divided into multiple square grids of the same size, with each grid representing a pixel. Simulations are then performed under the same process conditions for exposure of 1x1, 10x10, 5x20, and 100x100 rectangular beams, acquiring deposition energy data within each grid of the two-dimensional plane. The 1x1 beam is referred to as the pixel beam; 10x10 and 5x20 beams are considered small beams; and the 100x100 beam is considered a large beam. S102. Anomaly identification based on the extreme value-LOF algorithm: After completing the two-dimensional planar deposition energy data acquisition, for the deposition energy generated by each type of spot, firstly, the deposition energy data in each grid is traversed to perform 8-neighborhood extreme value detection, including maximum value detection and minimum value detection; then, for the detected extreme value data, other data in its 5x5 neighborhood are used to perform anomaly identification based on the LOF algorithm to determine whether the extreme value data is anomaly data; S103. Abnormal data replacement based on symmetry filtering: For each identified abnormal data, firstly, other deposition energy data within its surrounding 5x5 neighborhood are selected, and the abnormal data is removed. The remaining data is then subjected to mean filtering to obtain the neighborhood filter value. Subsequently, the symmetric coordinates of the abnormal data coordinates about each symmetry axis of the beam spot are calculated, and the neighborhood filter value of each corresponding grid is calculated. Finally, the average value of all neighborhood filter values ​​is calculated as the final filtering result, which is used for abnormal data replacement.

3. The electron beam deposition energy prediction method based on a convolution prediction model according to claim 2, characterized in that, The specific process of step S2 is as follows: S201. Binarization of graphic data: For each spot, set the value of the area covered by the spot in the two-dimensional plane to 1 and the value of the other areas to 0 to obtain the binarized graphic function corresponding to each spot. S202. Construct a deposition energy density function DEDF based on the selected point diffusion function PSF, which is used to calculate the deposition energy generated by electrons in each grid on the surrounding grid. S203. Based on the two-dimensional planar convolution formula, the deposition energy density function is convolved with the binarized graphic function corresponding to each spot to obtain the corresponding convolution prediction model.

4. The electron beam deposition energy prediction method based on a convolution prediction model according to claim 3, characterized in that, Deposition energy density function DEDF The expression is as follows: in K This is the proportionality coefficient. D The beam spot dose is the parameter to be fitted. K as well as PSF The parameter (x, y) is the position coordinate.

5. The electron beam deposition energy prediction method based on a convolution prediction model according to claim 4, characterized in that, The specific process of step S3 is as follows: S301. Parameter initialization based on pixel spot deposition energy distribution data and spot dose obtained in step S1: First, calculate the scaling factor K in the deposition energy density function using the pixel spot deposition energy distribution data and spot dose. Then, use the particle swarm optimization algorithm to initialize the parameters in the deposition energy density function. PSF The parameters are fitted to obtain the initial values ​​of the PSF parameters. Where N is the number of PSF parameters; S302. Based on the deposition energy data of the two small-sized spot sizes obtained in S1, the PSF parameter in the deposition energy density function is fitted to obtain the corrected values ​​of the PSF parameter in the deposition energy density function. S303. Using the deposition energy data of large-size beams obtained in S1, the optimal parameters are searched iteratively using a bisection method based on the initial and corrected values ​​of the PSF parameters to obtain the optimal fitting values ​​of each parameter.

6. The electron beam deposition energy prediction method based on a convolution prediction model according to claim 5, characterized in that, The specific process of step S302 is as follows: initialize the PSF parameters to the initial values ​​obtained from parameter initialization. Adding or subtracting 50% is used as the upper and lower limits for parameter correction. Then, using the deposition energy data of the two small-sized spot sizes obtained in S1, the PSF parameter in the deposition energy density function is fitted based on the particle swarm optimization algorithm within the upper and lower limits of parameter correction, and the corrected values ​​of the PSF parameter in the deposition energy density function are obtained respectively. (i=1,2).

7. The electron beam deposition energy prediction method based on a convolution prediction model according to claim 5, characterized in that, The specific process of step S303 is as follows: using the deposition energy data of the large-size spot obtained in S1, the average gradient of PSF parameter variation is calculated based on the two sets of PSF parameter correction values ​​and initial values. , and then with With the origin, along The direction is searched iteratively using a binary search method, with an initial step size of . The iterative search stops when the change in the objective function is less than the threshold in each search; the final search result is the optimal fit value of the PSF parameter in the deposition energy density function.

8. The electron beam deposition energy prediction method based on a convolution prediction model according to claim 5, 6, or 7, characterized in that, The proportionality coefficient K is calculated as follows: 。 Given a 2D plane with a grid of M pixels, where each pixel's coordinates are (x, y, y). i ,y i ), with an area of ​​S, and the deposition energy within each pixel grid is E(x). i ,y i The pixel spot dose is D0.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program, when run by a processor, performs the steps of the method as described in any one of claims 1-8.

10. An electron beam deposition energy prediction system based on a convolution prediction model, comprising an interconnected memory and a processor, wherein the memory stores a computer program, characterized in that, The computer program, when run by a processor, performs the steps of the method as described in any one of claims 1-8.

Images